Informs Annual Meeting 2017

TC29

INFORMS Houston – 2017

3 - Collusion in Markets with Syndication John Hatfield, McCombs School of Business, 2110 Speedway, Stop B6600, Austin, TX, 78172, United States, john.hatfield@gmail.com, Scott Kominers, Richard Lowery Markets for IPOs and debt issuances are syndicated, in the sense that a bidder who wins a contract may invite losing bidders to join a syndicate to fulfill the contract. We show that in such markets, standard intuitions from industrial organization can be reversed: Collusion may become easier as market concentration falls, and market entry may facilitate collusion. In particular, price collusion can be sustained by a strategy in which firms refuse to join the syndicate of any firm that deviates from the collusive price. Our results thus can rationalize the apparently contradictory facts that the market for IPOs exhibits seemingly collusive pricing despite its low level of market concentration. 4 - Signal Design and Costly Entry Vitor Farinha-Luz, ancouver School of Economics, 6000 Iona Drive, Vancouver, QC, Canada, vitor.farinhaluz@ubc.ca A buyer observes a noisy signal about his own value for a good before making a costly entry decision. The signal structure can be designed by the seller, who cannot commit to a posted price post-entry. Following entry, the buyer perfectly observes his valuation and the seller makes him an offer. I show that, if entry occurs with interior probability, the optimal information structure provides partial information and solves a trade-off between efficient entry and guaranteeing positive rents to the buyer.

simultaneously, the problem becomes even harder because of its combinatorial nature. In this paper, we present new flexible Bayesian models and efficient inference algorithms by taking both annotator reliability and label dependency into account. Extensive experiments on real-world datasets reveal that the proposed methods outperform other competitive alternatives.

TC30

350F Health Analytics with User-generated Data Invited: Social Media Analytics Invited Session Chair: Shuai Huang, University of Washington, University of Washington, Seattle, WA, 98195, United States, shuai.huang.ie@gmail.com Co-Chair: Xia Hu, PhD, Texas A&M University, College Station, TX, 77843, United States, hu@cse.tamu.edu 1 - Mental Health Monitoring using Dynamic Vocal Biomarkers Ying Lin, University of Houston, Houston, TX, 77030, United States, ylin58@uh.edu, Shuai Huang Modes disorder is inherently related to emotion. With recent progresses in linking the acoustic and linguistic features to emotion, automatic analysis of audio data could provide a powerful tool to assist in detection and monitoring of mode disorder. The audio data contains a rich array of features, such as the pitch, energy and linguistic content, that can inform or indicate signify emotion changes in the communication. This study employed a variety of signal processing techniques to convert the audio signals into dynamic vocal biomarkers that capture the acoustic and linguistic features in speech. We further monitored the underlying emotion based on the identified dynamic vocal biomarkers. 2 - Graph Theoretic Compressive Sensing Approach for Classification of Global Neurophysiological States from Electroencephalography Signals Prahalada Rao, University of Nebraska-Lincoln, Lincoln, NE, United States, rao@unl.edu Prahalada Rao, University of Nebraska-Lincoln, Lincoln, NE, 68588, United States, rao@unl.edu, Samie Tootooni, Chun-An Chou, Vladimir Miskovic We present a data-fusion framework integrating graph theoretic and compressive sensing (CS) techniques to detect global neurophysiological functions states using high-resolution electroencephalography (EEG) recordings. Acute stress experiments were induction (and control procedures) were used to elicit distinct states of neurophysiological arousal. Our experimental results revealed that the proposed graph theoretic compressive sensing approach yielded better classification performance (~90% F-score) compared to SVM use with statistical features (~50% F-Score). 351A Risk Analysis Contributed Session Chair: Sadegh Kazemi Zarkouei, Washington State University, Pullman, WA, United States, m.kazemi@wsu.edu 1 - Pricing Decisions when Consumers Have Access to Quality Reviews and a Secondary Market Sadegh Kazemi Zarkouei, Washington State University, Todd Hall Addition 470, Pullman, WA, 99164, United States, m.kazemi@wsu.edu, Michelle Wu We are studying the effect of strategic consumers’ behavior on a firm’s profit when consumers socially interact among each other in the primary and secondary market in two periods. When the firm introduces a new product to the primary market, it may face a group of strategic consumers in the first period who postpone their purchases with the hope of a discounted price in the second period. These forward-looking behaviors are exacerbated when consumers have the opportunity to socially interact with each other through quality reviews. The presence of social learning and the secondary market significantly influence the consumers’ purchasing decision and the firms optimal profit. TC31

TC29

350E Analytics for Social and Business Networks Sponsored: Artificial Intelligence Sponsored Session Chair: Kang Zhao, University of Iowa, Iowa City, IA, 52242, United States, kangzhao7@gmail.com 1 - Binary Matrix Completion Methods for Reciprocal Recommendation with Applications in Online Dating

Qihang Lin, University of Iowa, 21 East Market Street, S380, Pappajohn Business Building, Iowa City, IA, 52245, United States, qihang-lin@uiowa.edu, Kang Zhao, Xi Wang The success of an online dating site depends on its effectiveness in recommending potential partners to each other. Compared to the traditional recommendation systems, the recommendation systems used by online dating sites must predict the mutual interests of a male and a female users in each other, which is a reciprocal recommendation system. In this paper, we study various binary matrix completion methods that can integrate the different behavior information of male and female users to improve the overall prediction accuracy for reciprocal recommendation. The proposed method is based on solving non-convex matrix optimization problems by alternating proximal gradient methods. 2 - Understanding the Impact of Individual User’s Rating Patterns on Predictive Accuracy of Recommender Systems Jingjing Zhang, Indiana University, 1309 E.10th St, HH4143, Bloomington, IN, 47405, United States, jjzhang@indiana.edu, Xiaoye Cheng, Lu Yan This paper investigates how the rating patterns of individual users affect the performance of recommendation algorithms. We measure each individual user’s rating patterns from three different perspectives: rating value, rating structure and neighborhood network embeddedness. We study how these three groups of rating measures influence the recommender systems’ predictive accuracy performance for each user. Our experiments use five real-word datasets with varying characteristics. Our experimental results show consistent and significant effects of several rating measures on recommendation accuracy. 3 - Analyzing Firms’ Performance -A Supply Chain Network Perspective John Rios, University of Iowa, Iowa City, IA, 52242, United States, john-riosrodriguez@uiowa.edu, Kang Zhao, Jennifer Blackhurst This research explores the relationship between the performance of a firm and the structure of its supply chain and competition networks. The analysis starts with building the supply chain network and the competition network among 4.7k firms. Incorporating baseline features of firms’ individual characteristics, we show that firms’ structural features based on its supply chain network and competition network (e.g., diversity or coordination) contribute to the analysis of firms’ performance. 4 - Multi-label Annotation Aggregation in Crowdsourcing Xuan Wei, University of Arizona, Tucson, AZ, United States, weix@email.arizona.edu, Dajun Daniel Zeng, Junming Yin As a means of human-based computation, crowdsourcing has been widely used to annotate large-scale unlabeled datasets. One of the obvious challenges is how to aggregate these possibly noisy labels provided by a set of heterogeneous annotators. When each instance is associated with many possible labels

346